[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: initrd on installed kernels



"Christian T. Steigies" <cts@debian.org> writes:

> On Mon, Jun 28, 2004 at 02:00:00PM +0200, Goswin von Brederlow wrote:
>> Jeff Bailey <jbailey@raspberryginger.com> writes:
>> 
>> > On Thu, 2004-06-24 at 03:35, Thibaut VARENE wrote:
>> >
>> >> RE Jens' mail: the initrd used is the stock one, I didn't change
>> >> anything (yet).
>> >
>> > Please note that the initrd is generated when you install your kernel
>> > (or when you tell it to do so...)
>> >
>> >> I do agree that having tons of unused drivers builtin isn't a Good
>> >> Thing, but I was wondering whether some kind of compromise between
>> >> what we had (non-initrd, "quick" boot) and what we have (initrd,
>> >> "slow" boot) could be found.
>> >
>> > Can you say specifically where it's being slow?  Most of my boxes take
>> > about 80 seconds from the time it stops answering ping to the time that
>> > it starts again.  (So that's shutdown of filesystems, BIOS reboot, grub
>> > with its 5 second pause, kernel load, initrd load with a further 5
>> > second pause and then bootup to network).
>> >
>> > Having not only specific time differences to work from, but also where
>> > it's taking the time would be useful to know.  It really shouldn't spend
>> > a significant amount of time in the initrd at all.
>> >
>> > Tks,
>> > Jeff Bailey
>> 
>> Actually initrd should be faster than a monolithic kernel. The
>> monolithic kernel would try to initialize all the build-in drivers
>> which can take a significant time (like seconds) for some.
>
> On my notebook it takes ages to load the kernel image. With initrd it takes
> a lot longer to load, since initrd is so big.. Once the kernel is running,
> hdd access is much faster, but in the initial stage, it seems as slow as
> loading from a floppy. I guess it is just a configuration problem, but I
> haven't been able to convince the harddisk yet to load a little faster
> during boot.
>
> Christian

That is probably caused by the initrd being overly huge too.

Last one I checked was ~16MB big (uncompressed).


I'm testing if its possible to have the kernel autodetect attached
(just like the initrd can be attached to a kernel image pre/post build
on some archs) modules and insmod them inside the kernel before
mounting /.

That would save the extra space for a libc, shell, insmod, pivot_root,
chroot, mount + initrd script and would reduce it to one single file.
It would also simplify making a boot kernel for systems thats can't
boot a seperate initrd or post link it.

MfG
        Goswin



Reply to: